A Claude Code skill that helps you craft production-ready Seedance 2.0 (即梦) video prompts — from creative concept to copy-paste-ready prompt with storyboard.
- Full CRAFT framework: Context / Reference / Action / Framing / Tone
- Time-axis storyboarding for 10s+ videos with per-second breakdown
- Complete camera language library (dolly, orbit, pan, tilt, crane, POV...)
- Six core prompt principles baked in (positive-only, specific subjects, physics words, constraints)
- Auto quality checklist before output (10-point verification)
- Pitfall warnings (real-face rejection, MP3-only audio, file count limits)
- Long video production guidance (extension relay, segment editing, action relay)
- Understand your brief — purpose, footage, duration, aspect ratio, style, audience
- Creative concepts — proposes 2-3 directions with rationale
- Production prompt — outputs a complete prompt with time-axis storyboard, camera instructions, and constraint words
- Quality check — verifies against 10-point checklist before delivery
Option A: Plugin install (recommended)
/plugin marketplace add xueyuanhuang/seedance-prompt-skill
/plugin install seedance-prompt-skill@seedance-prompt-skill
Option B: Manual install
mkdir -p ~/.claude/skills/seedance
curl -o ~/.claude/skills/seedance/SKILL.md \
https://raw.githubusercontent.com/xueyuanhuang/seedance-prompt-skill/main/skills/seedance/SKILL.md/seedance 帮我做一个15秒的产品广告视频
/seedance I want a cinematic 10s video of a person walking into a futuristic city
Or just describe what you want naturally:
我有一张照片，想用 Seedance 做一个搞怪的短视频，7秒，竖屏
Personal intro video:
/seedance 用我的照片做一个15秒的自我介绍视频，科技感，紫色调
Product ad:
/seedance 帮我做一个手机产品展示视频，5秒，360度环绕，广告大片感
Fun social media clip:
/seedance 我想做一个搞怪的视频发朋友圈，从日常场景突然变成赛博朋克世界
| Feature | Spec |
|---|---|
| Input | Image (max 9) + Video (max 3) + Audio (max 3, MP3 only) + Text |
| Max files | 12 total, recommended 5-8 |
| Duration | 4-15 seconds |
| Aspect ratio | 16:9 / 9:16 / 1:1 |
| Resolution | 480p / 720p / 1080p |
| Modes | Text-to-video, First/Last frame, Multi-modal reference |
| Platform | jimeng.jianying.com |
- Chinese first — no need to translate, Chinese works natively
- Positive only — never use "don't" or "no", describe what you WANT
- Intent over detail — let AI handle execution details
- Specific subjects — describe precisely enough to picture it
- Physical motion words — add "缓慢", "连贯", "借助惯性" etc.
- Always add constraints — face stability, no distortion, smooth motion
MIT